EP0079251A2 - Geometrische Zeichenerkennung mit Darstellung des Skelettes und der Strichbreite - Google Patents

Geometrische Zeichenerkennung mit Darstellung des Skelettes und der Strichbreite Download PDF

Info

Publication number
EP0079251A2
EP0079251A2 EP82401438A EP82401438A EP0079251A2 EP 0079251 A2 EP0079251 A2 EP 0079251A2 EP 82401438 A EP82401438 A EP 82401438A EP 82401438 A EP82401438 A EP 82401438A EP 0079251 A2 EP0079251 A2 EP 0079251A2
Authority
EP
European Patent Office
Prior art keywords
character
code
processor
branches
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
EP82401438A
Other languages
English (en)
French (fr)
Other versions
EP0079251A3 (en
EP0079251B1 (de
Inventor
Pepe Siy
James Ching-Ting Young
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Unisys Corp
Original Assignee
Burroughs Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Burroughs Corp filed Critical Burroughs Corp
Publication of EP0079251A2 publication Critical patent/EP0079251A2/de
Publication of EP0079251A3 publication Critical patent/EP0079251A3/en
Application granted granted Critical
Publication of EP0079251B1 publication Critical patent/EP0079251B1/de
Anticipated expiration legal-status Critical
Expired - Lifetime legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/34—Smoothing or thinning of the pattern; Morphological operations; Skeletonisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries

Definitions

  • the invention relates to a character recognition system for use with document processing machines, and the like. It has utility for use with printed characters in many formats and for unconstrained handwritten characters. It employs a character code recognition technique, which may be called a geometric technique, making use of skeleton and stroke width data in digital form which is compared with data in standard character code patterns to enable identification. Character codes not identified initially by comparison with standard character codes are processed further and submitted to further tests to identify additional characters.
  • a character code recognition technique which may be called a geometric technique, making use of skeleton and stroke width data in digital form which is compared with data in standard character code patterns to enable identification. Character codes not identified initially by comparison with standard character codes are processed further and submitted to further tests to identify additional characters.
  • Prior art systems for document processors for recognizing printed characters in various formats and handwritten characters, such as the courtesy amounts on checks, include systems making use of statistical techniques.
  • One of the difficulties with such systems is that it is hard to adapt them to interpret representations of characters which were not included in the original statistical analysis.
  • the present system overcomes this difficulty very handily by enabling desired changes to be made in the memory forming a look-up table. This makes it possible to alter the selectivity and the sensitivity of the system in ways not readily available to the prior art.
  • a subspace technique has been implemented successfully for recognizing certain machine printed characters where the term "subspace" is applied because only a portion of the entire space that a character occupies is used to compare against prototype classes. It is natural that one would want to extend this technique to the recognition of other machine printed characters and to unconstrained handwritten characters considering the elegant simplicity of the technique. Unfortunately, the subspace technique has not proven to be adaptable to cope with the variables associated with some fonts of printed characters and with unconstrained handwritten characters for reasons set out below.
  • the subspace technique depends heavily upon the use of particular characters which can be segmented, isolated, or boxed before feature extraction is undertaken.
  • the characters must be formed with a constant pitch.
  • the subspace technique depends generally upon partitioning each character in a specific way such as into 3 x 3 or 5 x 5 equally divided cells.
  • the average gray scale in each cell serves as a feature, requiring a 9 feature vector for 9 cells and a 25 feature vector for 25 cells.
  • the distribution of the feature vectors in each class of characters (such as the character 3) is modelled by its auto-correlation, a second order model which corresponds to the mean and standard deviation in the case of one random variable.
  • This modelling has been adequate for some machine printed characters, since the position of the character stroke does not change very much from one character font to another.
  • Figure 2A shows some samples of unconstrained handwritten characters of class 3 (where each digit 0-9 is denoted as a class) against a background of 25 cells. In these examples, the stroke is seen to migrate drastically from one region to another of the 5 x 5 cells, demonstrating the inadequacy of auto-correlation modelling of the distribution of the feature vectors of an unconstrained character class.
  • the geometric technique employed in the present invention avoids problems associated with the subspace technique, since it does not require segmentation prior to feature extraction.
  • Character structures can be altered with ease for data representing noise codes (i.e. codes which are not recognized as character codes) and undesired structures can be removed with ease from such data to provide representations of a known character code.
  • Confusion between classes (such as between classes.4, 8 and 9 in Fig. 10A) can be resolved by using special test features to highlight their differences, using features extracted from the original resolution of the character image.
  • a character recognition system for use with document processing systems, and the like, in identifying characters, such as printed alpha-numeric characters of various formats, characters representing other symbols which are not alpha-numeric and unconstrained handwritten characters.
  • the system includes means for processing data received from outside the system in the form of signals representing selected geometric features of such characters, such as details concerning the'shape of the skeleton and stroke width characteristics.
  • the system includes means for extracting certain data representing features and branch classification from said signals and producing character codes representing the characters.
  • the system includes means for comparing the data representing the resulting unknown character codes with data representing known character codes. If, as a result of a comparison, a positive identification is made, the processor provides output signals designating the identified character.
  • the system transmits the data representing the features of the unknown character for further processing as unidentified data.
  • Additional processing means is coupled responsive to said unidentified data either to directly attempt an identification or to make an analysis thereof and to provide modified data incorporating information from said unidentified data and from said analysis. In the latter case, the modified data is then routed to said means for comparing unidentified data with known data in search of positive identification.
  • the additional processing means includes a "confused code" processor for applying fine feature tests including an analysis relating to stroke width to distinguish between different characters which are sufficiently alike to produce identical character codes. If the tests are successful, the confused code processor provides a readout identifying the character; if not successful, the processor diverts the character code for human interpretation.
  • a "confused code” processor for applying fine feature tests including an analysis relating to stroke width to distinguish between different characters which are sufficiently alike to produce identical character codes. If the tests are successful, the confused code processor provides a readout identifying the character; if not successful, the processor diverts the character code for human interpretation.
  • the additional processing means also includes a "noise processor" for removing unwanted elements, or noise, from, and sometimes concatenating, said data representing features.
  • the noise processor then provides modified data, less data representing unwanted elements, to said means for comparing in search of positive identification.
  • processing is indicated as taking place in a serial fashion. Because of the speed with which data is produced in an embodiment envolving the reading of many documents per record, processing in a preferred embodiment will actually be performed in a parallel look-ahead mode.
  • FIG. 1 For an overall showing of a system to which the present invention relates.
  • image data is applied over bus 12 from a reading 'and preliminary processing apparatus indicated at 10.
  • This image data is derived from scanning documents, such as a document at D, as they move by in a document transport track T in the direction of an arrow A.
  • An exemplary document transport system is shown in a copending U.S. Patent Application entitled “Controller for a Servo Driven Document Feeder", filed June 30, 1980 in the name of Harold A. Fasig, which has been assigned to the same assignee as the present invention and is hereby incorporated by reference.
  • the image data over bus 12 is analyzed by a Thinner and Stroke Width Analyzer at 14 which provides data in the form of digital signals representing the bare skeleton of each character together with digital signals representing stroke widths.
  • the data from the analyzer at 14 is supplied over bus 16 to a Feature Extractor or Branch Classifier at 18 which extracts gross features, or topological features, and produces character codes, or structure codes, including information as to topological features for each character detected by the reader and processor.
  • a character code, or structure code as the concatenation (or linkage in series) of all the related branch features of a character and their related nodes.
  • the feature extractor 18 in a preferred case may be implemented by the use of a computer such as a Burroughs B1900 or a larger computer, although hard logic or TTL Proms may be used, to extract topological features such as the number of branches, number of loops, types of branches, and the like and provide other operations necessary to generate character codes.
  • a computer such as a Burroughs B1900 or a larger computer, although hard logic or TTL Proms may be used, to extract topological features such as the number of branches, number of loops, types of branches, and the like and provide other operations necessary to generate character codes.
  • the character codes are supplied over bus 20 to a look-up table 22 which provides a comparison with data in the Prototype Structure Table (or template) 40 and classifies the received information according to whether there is (1) a direct match of the received character code with the data (or template) in the Prototype Table, (2) a "confused class” or (3) a "noise code". Descriptions relating to a (1) direct match of data, (2) a confused code or class and (3) a noise code are given below.
  • the look-up table 22 may be formed preferably by TTL hardwire logic.
  • the Prototype Structure Table 40 is preferably formed by a TTL Prom.
  • the Feature Extractor 18 is formed in a preferred case using TTL hardwire logic.
  • signals designating the identified character are transmitted over bus 24 and bus 26 directly to a bus at 28 for use as a direct read-out of the identified character.
  • This information may be supplied to utilization apparatus which may include a microprocessor and a printer.
  • the look-up table indicates that the character code represents more than one character or class of data, rather than a single character or class of data, a condition is indicated which may be labeled (2), a confused code or confused class.
  • a procedure involving special test features is required to resolve- the confusion. Suitable test features are provided by apparatus identified as the Confused Code Processor 32 which receives confused code signals over bus 30 from the look-up table 22 and bus 24. Processor 32 provides identification signals over bus 28. In case identification is not completed, it routes the identified data over bus 33 to the look-up table.
  • the Confused Code Processor will preferably be formed using a microprocessor such as an Intel 8035.
  • the character code from 18 is not in the code table of 22, meaning that it cannot be identified as belonging to either (1) or (2) above, we call it a noise code corresponding to class (3) above. In that case it is routed via busses 24 and 34 to the Noise Processor 36 for processing and then over bus 37 for comparison in the look-up table.
  • the Noise Processor will preferably be formed using a microprocessor such as an Intel 8035.
  • Block 40 is a source of prototype structure information in the form of a table which may be altered by additions or deletions as needed for access over a bus 38 from the look-up table at 22. This demonstrates how the Prototype Structure Table may be modified by the simple addition of instructions to a TTL Prom, or other memory device or the subtraction of instructions from such a memory device.
  • the Reader and Processor 10 and the Thinner and Stroke Width Analyzer 14 are disclosed in detail in other applications which are referenced herein as a source of signals to be employed in the practice of this invention.
  • Examples of a Reader and Processor are shown in a copending United States patent application number 35,993 (now abandoned) filed in the names of Pepe Siy, Robert Bloss and Gary B. Copenhaver, on May 4, 1979 entitled “Adaptive OCR Front-end System” which is assigned to the same assignee as the present application.
  • Another example of a Reader and Processor is to be found in a copending U.S. Patent application Number (docket 21-1402) entitled “Pixel Correction Logic Circuit” filed in the names of Jimmie Neill and David Sutton.
  • the new skeleton representation puts forward the basic structure of the character which may be used, in a case (1) example, in direct identification of the character. Use of the skeleton may be made also to facilitate the extraction of geometric features and the further analysis needed to resolve confusion in cases (2) and (3) above.
  • the skeleton representation facilitates the description of the character structure.
  • Each of these "points" is determined by looking at the skeleton points associated with a point centered in a three-point-by-three point window. If the central point has only one neighbor (Fig. 3C), then the central point is an end point. If the central point has two neighbors (Fig. 3D), then it is a branch point. If the central point has three or more neighbors (Fig. 3E), then it is a junction point.
  • the end points and junction points may be referred to as critical points which may also be called node points.
  • the critical points, or node points, are numbered in a particular sequence in the practice of this invention. It is recognized, of course, that the nodes could be numbered in various arbitrary ways, but for purposes of this invention there is only one preferred way. That way consists with vertical features of always designating the first critical point, or node, as the top and the second critical point as the bottom. With horizontal features, the first node is always the left node and the second is always the right node. The convention can thus be summarized by saying the first critical point is at the top or left and the second critical point is at the bottom or right with clustered nodes or connected nodes given the same label. The application of this convention is illustrated in Fig. 9.
  • a collection of end points and/or junction points defines a node set, which may be called node 1, node 2 etc.
  • a branch therefore, may be said to be formed by a sequence of branch points starting from one critical point (node) and ending at another (or the same) critical point (node).
  • a collection of branches defines a branch set.
  • a node set and/or a branch set may be used to define the structure of a character.
  • (Nl, N2) defines a node set
  • (Bl, B2) defines a branch set
  • (Nl, N2) (Bl, B2) defines the structure of the character "9".
  • the character "1", as shown in Fig. 9, may be defined by a node set Nl, N2.
  • each branch is classified by the Feature Extractor 18 in a gross, or rough topological fashion, to establish conformation to one of the following classes: line, curve, or loop.
  • Three measurements are used to establish this classification. They are index of modulation (DM), least square error(S), and branch length (MD).
  • the index of modulation may be defined as the ratio of the length of the straight line joining the two nodes of a branch to the length of the branch joining the two nodes. By this definition the index of modulation of a straight line is 1, of a closed loop is 0, and of a curve is between 0 and 1.
  • FIG. 4A, 4B and 4C illustrate the index of modulation (DM) as the relationships of a straight line Ll to a curve L2, a line of zero length to a loop and a line Ll to a line L3.
  • Figure 5 is a flow chart illustrating how this gross classification is accomplished working with data in the form of digital representations of DM, S and MD.
  • branches may be classified further into the following sets or features where it will be recognized that additional and/or other sets or features may be used and still be within the scope of the invention:
  • Fig. 6A Examples of these are illustrated in Fig. 6A.
  • the character structure i.e. the connectivity of the character branches through their nodes, becomes explicit in the skeleton representation.
  • a character code is defined. To generate the code each branch is assigned to one of the sets or features illustrated in Fig. 6A.
  • Fig. 6A The features of Fig. 6A were selected to represent the numerical character classes. Each of the features defines an equivalent class of allowable distorion in a branch. For example, the branches shown in Fig. 6B should all be classified as D shaped curves as shown in column D opposite "Pictorial Representation" in Fig. 6A. It will be recognized that the characters of interest could include characters of many other types than just the numerals illustrated, such as the characters represented in the ASCII code, without departing from the spirit of the invention.
  • a membership grade or "degree of belongingness" may be defined for each feature of line, curve or loop.
  • the membership grade of a line feature is determined by its slope angle.
  • a threshold angle O T is selected which establishes a boundary between the horizontal and vertical lines where 0 is the angle of the line relative to the horizontal.
  • the membership grades for a line may then be defined as:
  • the membership grade of one of the curve features is determined by partitioning the curve branch into 3 equal segments where, for each curve feature, a preferred direction is defined on each segment.
  • the membership grade of a curve feature is obtained then by counting the number of branch points that satisfy the preferred direction, normalized to the number of points in each segment.
  • Figure 7 shows the preferred direction for an idealized C curve.
  • Fig. 16 illustrates an exemplary D curve which has been chosen at random to illustrate the method of segmentation used.
  • Two vectors IXD and IYD are defined as: where IX and IY represent the X and Y branch coordinate lists.
  • the +1 and -1 entries in IXD mean right and left movement across the branch, respectively.
  • the +1 and -1 entries in IYD mean down and up movement across the branch, respectively.
  • Two counting vectors IXR and IXT are employed in the system.
  • Vector IXR involves a count of the number of right, left, up, and down movements in each segment.
  • Vector IXT involves a count of the number of entries in each segment.
  • Each feature template is defined by a vector with a nonzero entry in the preferred direction in each segment.
  • Figures 17 and 18 show the counting vectors IXR, IXT and Figure 19 shows examples of template vectors.
  • Rx The correlation between the preferred motion of the branch given by IXR and the preferred motion of each feature defined by its template X is given by Rx, as follows:
  • the membership grade of a loop feature is determined by the relative position of the.node with respect to the top, Y top, and the bottom, Y bot boundary of the branch.
  • the position threshold Yth is selected between Y top and Y bot, see Figure 8.
  • the membership grades are:
  • the branch features are used to describe the structure of the characters. Two branches are said to be related if they have a common node.
  • a character code (structure code) is the concatenation, or linkage in a series, of all the related branch features and their corresponding node pairs.
  • Figure 9 shows typical Character codes for the numerical characters.
  • the character codes define equivalent classes of characters in a manner to include allowable distortions. New character codes may easily be added as new structures are found in the future and archived or deleted as their occurrence diminishes in frequency.
  • a table of allowable character codes is generated and stored in the look-Up table at 22. The codes stored in the table are unique representations of the character codes. First level character recognition is achieved by simple table lookup, that is structure matching.
  • Case (1) a direct match
  • Case (2) a confused class or code
  • Case (3) a noise code
  • the character code represents a unique class and recognition is complete.
  • case (2) the character code represents several classes which may be called a confused code and special test features are employed to resolve the confusion.
  • case (3) the character code is not in the code table and the code may be called a noise code.
  • Each branch is represented by coding the x, y coordinate of each consecutive branch point connecting a pair of nodes.
  • the x, y coordinate lists are packed together into one x, y coordinate list array.
  • the array is then delimited by a pointer array which contains an expression denoting the length of each branch.
  • Figures 10B-1 and 10B-2 show the skeleton representation of the 3 characters shown in Figures 10A-1 and 10A-2. The following information is stored in a test example, although, of course, ICL and IAUT would normally not be available at this stage from processing an unknown character code:
  • the first record in Figures 10B-1 and 10B-2 contains the following information:
  • the first record in Figure 10C contains the following information:
  • Figures 10A-1 and 10A-2 show three different characters 4, 8, 9 which, when analyzed in accordance with the present system, yield an identical character code, i.e. V(12) ° 0A(11), a case (2) situation involving a confused code or class. Special tests, involving fine features, are used by the Confused Code Processor indicated at 32 (Fig. 1) to resolve the confusion presented by this situation. As an example, the following features are used to resolve the confusion involved in distinguishing between the 4, 8 and 9:
  • the semantic part of the classification defines a piecewise linear region whose boundaries are defined by the thresholds. If the features are properly designed this.region will be well separated. The threshold parameters will be very stable and the determination not very critical.
  • each branch feature there are only a finite number of characteristics that one can extract.
  • Some typical features are length, average stroke width, stroke width standard deviation, aspect ratio, 5 directional chain code (where a branch is subdivided into five segments instead of three in order to increase the resolution), 5 average stroke width, acuteness or pointedness of a loop branch, etc.
  • These special test (fine) features can be precomputed so that, once a confused case (2) is encountered, the features are available to be used in the classification of each such confused class.
  • a noise code of case (3) above, is a code that is not in the code table.
  • noise of type (b), above is primarily caused either by individual writing style or by pre-processing.
  • Individual style refers to the way a person writes which results in such things as added loops and extended branches as indicated at "a" in Figure 11.
  • Pre-processing problems arise due either to an improper choice of threshold (Fig. 12) or to improper smoothing (Fig. 13).
  • Noise problems resulting from an improper choice of threshold level can cause: (a) a pin hole in a character stroke, resulting in a small bubble after the erosion or skeletonization process has been applied; (b) unwanted branches due to branch bridging or partial filling of loops creating multiple branches inside a loop; (c) removal of a necessary branch due to uneven gray level intensities along the strokes of a character.
  • Figure 12 shows some effects of improper thresholding. Improper smoothing, on the other hand, can produce rugged edges which appear as undesirable extraneous short branches, as indicated in Figure 13.
  • a noise code processing flow chart is shown in Figure 14: The algorithm is described in the following:
  • Fig. 14B is a chart depicting relationships between operations in accordance with the algorithm referred to above, various samples of skeletons of "noisy" and unidentifiable characters, the same characters following operations and reference characters with which identification is to be made.
  • Figure 20A is a representation of a series of handwritten characters as they appeared on a document to be read.
  • Figure 20B is a representation of the skeletons of the same characters as they appeared on a cathode ray tube after processing through reader and processor 10 and thinning in the thinner and stroke width analyzer 14. It was found that recognition of all but the "5" and the "8” could be made with 100% confidence in the result, using only the look-up table 22 and the associated prototype structure table 40. Confidence in the result of direct recognition for the "5" is about 83% and for the "8” is about 91%. Recognition of the "5" and "8” in this case would approach 100% with additional processing in the noise processor 36.
  • Figure 21A is a representation of printed characters as they appeared on a document to be read.
  • Figure 21B is a representation of the skeletons of the same characters after processing through the reader and processor 10 and after thinning in the thinner and stroke width analyzer 14. In this example, as in other examples employing other printer character fonts, it was found that confidence in the result for nearly every character is 100%.
  • a technique is disclosed to enable recognition of unconstrained handwritten characters and also of printed characters in various formats.
  • Each character is first described in terms of its skeleton and the associated stroke width information of each point in the skeleton.
  • the skeleton is used to extract the structure of the character (synctatic information) represented by the character code, and the stroke width information is used to extract special test features (semantic information) represented by the discriminant function.
  • Class classifiers are used consisting of a collection of known character codes for a given class and their associated discriminant functions.
  • the classifier for class "x" is represented by: where N is the number of prototypes or styles in class "x".
  • the discriminant function is empty when the character code is unique, that is, when the structure of the character uniquely defines the character class.
  • the classifier representation provides an easy means of adding or subtracting character codes and their associated discriminant functions. An incoming unknown character is assigned to the class that contains the character code that best matches the character code of the unknown character and its associated discriminant function which yields the best response.
  • the skeleton and stroke width information representation of any character is complete in the sense that the original character can be reconstructed from them.
  • the representation provides us with a means of readily extracting gross geometrical or topological features such as the number of branches, number of loops, types of branches and their connectivity.
  • the representation also provides us with a means of extracting finer features by such techniques as comparing regional stroke width within a branch or between branches, and local shape features.
  • the ability to extract the needed information from gross to fine features about a character is one of the unique characteristics of the disclosed geometric recognition technique.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Character Discrimination (AREA)
  • Character Input (AREA)
EP82401438A 1981-09-11 1982-07-30 Geometrische Zeichenerkennung mit Darstellung des Skelettes und der Strichbreite Expired - Lifetime EP0079251B1 (de)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US30111781A 1981-09-11 1981-09-11
US301117 1981-09-11

Publications (3)

Publication Number Publication Date
EP0079251A2 true EP0079251A2 (de) 1983-05-18
EP0079251A3 EP0079251A3 (en) 1985-01-16
EP0079251B1 EP0079251B1 (de) 1990-01-31

Family

ID=23162028

Family Applications (1)

Application Number Title Priority Date Filing Date
EP82401438A Expired - Lifetime EP0079251B1 (de) 1981-09-11 1982-07-30 Geometrische Zeichenerkennung mit Darstellung des Skelettes und der Strichbreite

Country Status (3)

Country Link
EP (1) EP0079251B1 (de)
JP (1) JPS5848183A (de)
DE (1) DE3280099D1 (de)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0322920A2 (de) * 1988-01-04 1989-07-05 Sumitomo Electric Industries Limited Optische Zeichenlesevorrichtung
EP0335696A2 (de) * 1988-03-29 1989-10-04 Kabushiki Kaisha Toshiba Mustererkennungsgerät
EP0381773A1 (de) * 1988-07-20 1990-08-16 Fujitsu Limited Buchstabenerkennungsvorrichtung
DE4017430A1 (de) * 1989-05-31 1990-12-06 Hitachi Ltd Verfahren und vorrichtung zum erfassen von mustern
US4987603A (en) * 1988-09-22 1991-01-22 Glory Kogyo Kabushiki Kaisha Recognition method for letter patterns
EP0600585A2 (de) * 1992-11-20 1994-06-08 Mutoh Industries Ltd. System zum Identifizieren von Freihandzeichnungen
DE4308245C1 (de) * 1993-03-16 1994-11-03 Friedemann Groh Verfahren zur Erkennung von Schriftzeichen oder Schriftzügen
US5625710A (en) * 1988-07-20 1997-04-29 Fujitsu Limited Character recognition apparatus using modification of a characteristic quantity
EP2572317A4 (de) * 2010-08-30 2017-05-17 Alibaba Group Holding Limited Erkennung digitaler bilder

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH01175847A (ja) * 1987-12-30 1989-07-12 Taihei Sansho:Kk 治療用シート
JPH0358759U (de) * 1989-10-06 1991-06-07
JPH04104894U (ja) * 1991-02-15 1992-09-09 コンデ株式会社 体温縫いぐるみ

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4015239A (en) * 1974-10-21 1977-03-29 Hitachi, Ltd. Character recognition apparatus

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS5341136A (en) * 1976-09-28 1978-04-14 Agency Of Ind Science & Technol Pattern reading system
JPS5854432B2 (ja) * 1978-02-24 1983-12-05 日本電信電話株式会社 パタ−ン認識処理方法

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4015239A (en) * 1974-10-21 1977-03-29 Hitachi, Ltd. Character recognition apparatus

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS, vol. 4, no. 6, pages 570-575, IEEE, New York, US; P. SIY et al.: "Fuzzy logic for handwritten numeral character recognition" *
THE MARCONI REVIEW, vol. 32, no. 172, 1969, pages 3-20, Great Baddow, GB; K.H. HOSKING et al.: " A feature detection method for optical character recognition" *

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0322920A2 (de) * 1988-01-04 1989-07-05 Sumitomo Electric Industries Limited Optische Zeichenlesevorrichtung
EP0322920A3 (de) * 1988-01-04 1991-11-06 Sumitomo Electric Industries Limited Optische Zeichenlesevorrichtung
EP0335696A2 (de) * 1988-03-29 1989-10-04 Kabushiki Kaisha Toshiba Mustererkennungsgerät
EP0335696A3 (de) * 1988-03-29 1991-08-21 Kabushiki Kaisha Toshiba Mustererkennungsgerät
US5625710A (en) * 1988-07-20 1997-04-29 Fujitsu Limited Character recognition apparatus using modification of a characteristic quantity
EP0381773A1 (de) * 1988-07-20 1990-08-16 Fujitsu Limited Buchstabenerkennungsvorrichtung
EP0381773A4 (en) * 1988-07-20 1993-05-19 Fujitsu Limited Character recognition apparatus
US4987603A (en) * 1988-09-22 1991-01-22 Glory Kogyo Kabushiki Kaisha Recognition method for letter patterns
DE4017430A1 (de) * 1989-05-31 1990-12-06 Hitachi Ltd Verfahren und vorrichtung zum erfassen von mustern
EP0600585A3 (de) * 1992-11-20 1994-11-02 Mutoh Ind Ltd System zum Identifizieren von Freihandzeichnungen.
EP0600585A2 (de) * 1992-11-20 1994-06-08 Mutoh Industries Ltd. System zum Identifizieren von Freihandzeichnungen
DE4308245C1 (de) * 1993-03-16 1994-11-03 Friedemann Groh Verfahren zur Erkennung von Schriftzeichen oder Schriftzügen
EP2572317A4 (de) * 2010-08-30 2017-05-17 Alibaba Group Holding Limited Erkennung digitaler bilder

Also Published As

Publication number Publication date
EP0079251A3 (en) 1985-01-16
JPH0440749B2 (de) 1992-07-06
EP0079251B1 (de) 1990-01-31
DE3280099D1 (de) 1990-03-08
JPS5848183A (ja) 1983-03-22

Similar Documents

Publication Publication Date Title
KR100248917B1 (ko) 패턴인식장치및방법
Hanmandlu et al. Fuzzy model based recognition of handwritten numerals
Yu et al. A generic system for form dropout
Lai et al. Detection of dimension sets in engineering drawings
US5774580A (en) Document image processing method and system having function of determining body text region reading order
CA1299292C (en) Character recognition algorithm
US5787194A (en) System and method for image processing using segmentation of images and classification and merging of image segments using a cost function
KR100249055B1 (ko) 문자인식장치및방법
Shijian et al. Script and language identification in noisy and degraded document images
EP0079251B1 (de) Geometrische Zeichenerkennung mit Darstellung des Skelettes und der Strichbreite
Bushofa et al. Segmentation and recognition of Arabic characters by structural classification
Lehal et al. Feature extraction and classification for OCR of Gurmukhi script
Elgammal et al. A graph-based segmentation and feature extraction framework for Arabic text recognition
JP2007058882A (ja) パターン認識装置
Lam et al. Reading newspaper text
den Hartog et al. Knowledge-based segmentation for automatic map interpretation
JP3917349B2 (ja) 文字認識結果を利用して情報を検索する検索装置および方法
Al-Sadoun et al. A new structural technique for recognizing printed Arabic text
JP3095069B2 (ja) 文字認識装置、学習方法および文字認識プログラムを記録した記録媒体
Srinivas et al. An overview of OCR research in Indian scripts
JP4176175B2 (ja) パターン認識装置
Joshi et al. Combination of multiple image features along with KNN classifier for classification of Marathi Barakhadi
EP0684576A2 (de) Verbesserungen der Bildverarbeitung
Dhandra et al. On Separation of English Numerals from Multilingual Document Images.
Sahu et al. Survey and analysis of devnagari character recognition techniques using neural networks

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

AK Designated contracting states

Designated state(s): BE DE FR GB IT NL

17P Request for examination filed

Effective date: 19830416

PUAL Search report despatched

Free format text: ORIGINAL CODE: 0009013

AK Designated contracting states

Designated state(s): BE DE FR GB IT NL

17Q First examination report despatched

Effective date: 19860228

RAP1 Party data changed (applicant data changed or rights of an application transferred)

Owner name: BURROUGHS CORPORATION (A DELAWARE CORPORATION)

R17C First examination report despatched (corrected)

Effective date: 19870504

GRAA (expected) grant

Free format text: ORIGINAL CODE: 0009210

AK Designated contracting states

Kind code of ref document: B1

Designated state(s): BE DE FR GB IT NL

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: NL

Effective date: 19900131

Ref country code: IT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT;WARNING: LAPSES OF ITALIAN PATENTS WITH EFFECTIVE DATE BEFORE 2007 MAY HAVE OCCURRED AT ANY TIME BEFORE 2007. THE CORRECT EFFECTIVE DATE MAY BE DIFFERENT FROM THE ONE RECORDED.

Effective date: 19900131

Ref country code: BE

Effective date: 19900131

REF Corresponds to:

Ref document number: 3280099

Country of ref document: DE

Date of ref document: 19900308

ET Fr: translation filed
NLV1 Nl: lapsed or annulled due to failure to fulfill the requirements of art. 29p and 29m of the patents act
PLBE No opposition filed within time limit

Free format text: ORIGINAL CODE: 0009261

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: NO OPPOSITION FILED WITHIN TIME LIMIT

26N No opposition filed
REG Reference to a national code

Ref country code: FR

Ref legal event code: CD

ITTA It: last paid annual fee
PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: GB

Payment date: 19930611

Year of fee payment: 12

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: FR

Payment date: 19930706

Year of fee payment: 12

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: DE

Payment date: 19930728

Year of fee payment: 12

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: GB

Effective date: 19940730

GBPC Gb: european patent ceased through non-payment of renewal fee

Effective date: 19940730

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: FR

Effective date: 19950331

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: DE

Effective date: 19950401

REG Reference to a national code

Ref country code: FR

Ref legal event code: ST